OpenAI, Anthropic, and the big cloud providers are shoveling free compute and token credits at early-stage startups, and the numbers have gotten large enough that founders are treating them less like a discount and more like a funding round. The Wall Street Journal report that kicked this off pegs the offers to Y Combinator companies at $500,000 apiece. Some founders are stacking more than that.
How the bidding got here
Back in May, Sam Altman announced $2 million in token credits for a YC cohort, with an equity stake attached via uncapped SAFE agreements. Anthropic saw an opening and jumped in with $500,000 in Claude credits and no equity required, which, if you believe the reporting, is a 16x leap from its previous standard offer of around $30,000. Then OpenAI matched the no-equity $500K and bolted on an optional $1.5 million more for founders willing to give up shares.
So you can now take OpenAI's money two ways: clean, or with strings. The two-tier structure is the interesting part, because it lets OpenAI look as founder-friendly as Anthropic while still dangling the quasi-investor deal for anyone who wants the bigger check.
Hans Ibarra, who runs AI voice startup Dialogus, told the Journal his competing offers added up to more than $3 million in cloud and token credits. PitchBook puts the average US seed round at roughly that same size, which is the whole point. When a credit package equals your seed, you're not comparing model quality anymore. You're picking an infrastructure partner the way you'd pick a lead investor.
The cloud guys want in too
This isn't only a frontier-lab fight. A Google Cloud spokesperson said the company gives out up to $500,000 in credits, early access to Gemini models, and occasional time with DeepMind engineers. Microsoft and AWS run comparable programs. Cursor, the AI coding firm now owned by SpaceX, ran a 75 percent discount through July 5.
Worth keeping some skepticism about the totals here. The specific dollar figures move month to month, and a lot of what circulates comes from tracker sites and founder self-reports rather than anything the labs disclose directly. The $3 million number is one founder's account of competing offers, not a signed term sheet made public.
Why they're eating the cost
The math behind giving away millions is switching cost. Build a product on one model, wire your prompts and workflows around its quirks, burn through $500,000 doing it, and migrating to a competitor becomes a genuine engineering headache. The labs are betting that credits today turn into paying enterprise contracts once the free tier runs dry.
There's an awkward timing wrinkle. Both OpenAI and Anthropic are widely expected to go public, and generous free credits are exactly the kind of thing that dents gross margins and net revenue retention, the metrics IPO investors pick apart. The bet is that credit-seeded customers convert before the S-1 roadshow. Whether that conversion actually happens is the question nobody can answer yet, and one recent tally already puts the two labs' combined share of AI startup revenue near 89 percent, so there isn't much room left to grab.
For founders, the read is simpler. Run a competitive credit process before signing anything, because the labs have shown they'll pay seed-round money for a customer. Just remember the retail bill shows up in month 13 whether or not the product's unit economics are ready for it. With four YC cohorts a year of roughly 200 companies each, OpenAI and Anthropic together could spend up to $800 million on credits annually. The next cohort's offers will tell you whether the ceiling keeps climbing.
